U.S. patent number 5,067,162 [Application Number 06/879,987] was granted by the patent office on 1991-11-19 for method and apparatus for verifying identity using image correlation.
This patent grant is currently assigned to Identix Incorporated. Invention is credited to Edward C. Driscoll, Jr., Craig O. Martin, Kenneth Ruby, James J. Russell, John G. Watson.
United States Patent |
5,067,162 |
Driscoll, Jr. , et
al. |
November 19, 1991 |
Method and apparatus for verifying identity using image
correlation
Abstract
A method and apparatus for verification of personnel identity by
correlation of fingerprint images. The method includes the steps
of: first enrolling a person by the steps of forming a reference
image of a fingerprint of the person, identifying a plurality of
reference sections within the reference image, where the image data
contained in each of the reference sections is distinct relative to
the image data adjacent to and surrounding the reference section,
and saving the image data of each of the reference sections; and
then verifying the identity of someone claiming to be an enrolled
person by the steps of retrieving the image data of the reference
sections of the enrolled person, forming a verify image of the
fingerprint of the person claiming to be enrolled, where the verify
image includes a plurality of verify regions each corresponding in
position to one of the reference sections and each larger in extent
than its corresponding reference section, determining a best-match
location within each verify region at which the image data is most
similar to the image data of its corresponding reference section,
and verifying the identity of the person claiming to be enrolled
according to the degree of similarity between the image data of the
best-match locations and the corresponding reference sections and
according to the degree of similarity between the relative
positioning of the best-match locations and the corresponding
reference sections.
Inventors: |
Driscoll, Jr.; Edward C.
(Portola Valley, CA), Martin; Craig O. (Menlo Park, CA),
Ruby; Kenneth (Florence, OR), Russell; James J.
(Mountain View, CA), Watson; John G. (Menlo Park, CA) |
Assignee: |
Identix Incorporated
(Sunnyvale, CA)
|
Family
ID: |
25375295 |
Appl.
No.: |
06/879,987 |
Filed: |
June 30, 1986 |
Current U.S.
Class: |
382/126; 382/209;
382/278 |
Current CPC
Class: |
G07C
9/37 (20200101); G06K 9/001 (20130101); G06K
9/6211 (20130101); G06K 9/00067 (20130101) |
Current International
Class: |
G06K
9/00 (20060101); G07C 9/00 (20060101); G06K
009/62 () |
Field of
Search: |
;382/4,5,21,8,30,34 |
References Cited
[Referenced By]
U.S. Patent Documents
Foreign Patent Documents
|
|
|
|
|
|
|
83102997 |
|
Oct 1983 |
|
EP |
|
85108402 |
|
Jan 1986 |
|
EP |
|
Other References
Schiller, "Fingerprint Identification System", Fingermatrix Inc.,
Published EPA #0090377, 10/5/83. .
S. Kashioka et al., "Automatic Template Selection Technique for the
Local Pattern Matching Method", Systems & Computer in Japan,
vol. 17, No. 5, 1986; pp. 25-35. .
M. Schiller et al., "Fingerprint Verification Method", EPA 0 125
532; Ser. No. 84104529.7 filed Apr. 21, 1984. .
A. Shimizu, "Fingerprint Collator", Ser. No. 58-123186. .
Y. M. Ting et al., "Fingerprint Image Enhancement System", IBM
Technical Disclosure Bulletin, vol. 16, No. 8, Jan. 1974..
|
Primary Examiner: Boudreau; Leo H.
Attorney, Agent or Firm: Limbach, Limbach & Sutton
Claims
What is claimed is:
1. An image correlation method for verifying the identity of an
object, said method comprising the steps of:
forming a two-dimensional reference image of a reference
object;
identifying a plurality of two-dimensional reference sections from
said reference image by the steps of partitioning said reference
image into a plurality of two-dimensional candidate reference
sections from which said reference sections will be selected,
performing an autocorrelation calculation for each candidate
reference section to determine the degree of distinctiveness of
said candidate reference section relative to a two-dimensional
local region of said reference section by comparing pixels of the
two-dimensional image data of said candidate reference section to
corresponding pixels of the two-dimensional image data of
equal-sized areas throughout said local region, wherein said local
region is larger than said candidate reference section and smaller
than said reference image, and selecting said reference sections
from the most distinctive of said candidate reference sections
based on the results of the autocorrelation calculations; and
then
forming a two-dimensional verify image of an object whose identity
is to be verified, wherein said verify image includes a plurality
of two-dimensional verify regions each corresponding in position to
one of said reference sections and each larger in extent than its
corresponding reference section;
determining a best-match location within each verify region at
which the image data is most similar to the image data of its
corresponding reference section;
determining the relative positioning of said best-match locations
relative to each other and comparing that positioning to the
relative positioning of the reference sections relative to each
other; and
verifying the identity of said object according to the degree of
similarity between the image data of said best-match locations and
said corresponding reference sections, and according to the degree
of similarity between the relative positioning of said best-match
locations and the relative positioning of said corresponding
reference sections.
2. A method as recited in claim 1 wherein said step of forming a
two-dimensional reference image of a reference object includes the
steps of first forming a two-dimensional image consisting of grey
pixels having a plurality of brightness values on a grey scale
ranging between black and white, and then converting said grey
pixels into binary pixels having brightness values of black or
white.
3. A method as recited in claim 2 wherein said step of converting
said grey pixels into binary pixels includes the steps of first
determining a local median grey value of a subfield of said
reference image, and then converting each grey pixel within said
subfield into a binary pixel according to its grey scale brightness
value relative to said median grey value.
4. A method as recited in claim 1 wherein said step of performing
an autocorrelation calculation for each candidate reference section
includes the steps of computing a plurality of correlation values,
wherein each correlation value is based on a comparison of the
image data of said candidate reference section with the image data
of an equal-sized area of said local region, and computing an
autocorrelation score based on the highest correlation value
computed within said local region but outside of a region near the
location of said candidate reference section, wherein a high
autocorrelation score indicates a locally non-distinct candidate
reference section.
5. A method as recited in claim 4 wherein said step of computing a
correlation value includes the step of a pixel-by-pixel comparison
between the image data of said candidate reference section and the
image data of said equal-sized area of said local region.
6. A method as recited in claim 4 wherein said step of forming a
reference image of a reference object includes the step of forming
an image consisting of grey pixels having a plurality of brightness
values on a grey scale ranging between black and white, and
calculation includes the step of converting the grey pixels of said
candidate reference section and of said local region into binary
pixels having brightness values of black or white prior to
performing said step of computing correlation values.
7. A method as recited in claim 6 wherein said step of converting
the grey pixels of said local region into binary pixels includes
the steps of first determining a median grey value within said
local region, and then converting each grey pixel of said local
region into a binary pixel according to its grey scale brightness
value relative to said median grey value.
8. A method as recited in claim 6 wherein said step of converting
the grey pixels of said candidate reference section into binary
pixels includes the step of trinarizing the image data of said
candidate reference section by first determining a black-grey
threshold value and a grey-white threshold value based on the
distribution of grey scale brightness values within said candidate
reference section, and then converting the grey scale pixels of
said candidate reference section into black, grey, and white pixels
according to their grey scale brightness values relative to said
black-grey and grey-white threshold values.
9. A method as recited in claim 8 wherein said step of computing a
correlation value includes the step of comparing the black pixels
and white pixels of said candidate reference section to the black
pixels and white pixels of said local region, adding the number of
matches between black or white pixels of said candidate reference
section and black or white pixels of said local region, and
dividing by the number of black and white pixels of said candidate
reference section, wherein the resulting value equals said
correlation value.
10. A method as recited in claim 8 wherein said black-grey and
grey-white threshold values are selected such that about one third
of the grey scale pixels of said candidate reference section are
converted into black pixels and about one third of the grey scale
pixels of said candidate reference section are converted into white
pixels.
11. A method as recited in claim 1 wherein said step of
partitioning said reference image defines a plurality of candidate
reference sections that overlap adjacent candidate reference
sections, and wherein said step of selecting said reference
sections from the most distinctive of said candidate reference
sections includes the step of rejecting candidate reference
sections that substantially overlap adjacent candidate reference
sections having higher degrees of local distinctiveness.
12. A method as recited in claim 1 wherein said step of forming a
two-dimensional reference image of a reference object includes the
step of forming an image consisting of grey pixels having a
plurality of brightness values on a grey scale ranging between
black and white, and wherein said method further comprises the step
of rejecting certain candidate reference sections based on the grey
scale values of the pixels within said candidate reference
sections.
13. A method as recited in claim 12 wherein said step of rejecting
candidate reference sections includes the step of rejecting a
candidate reference section based on the median grey value within
said candidate reference section.
14. A method as recited in claim 13 wherein said step of rejecting
candidate reference sections includes the step of rejecting a
candidate reference section if said candidate reference section is
adjacent to a candidate reference section that has been rejected
based on its median grey value.
15. A method as recited in claim 12 wherein said step of rejecting
candidate reference sections includes the steps of first
determining a black-grey threshold value and a grey-white threshold
value based on the distribution of grey scale brightness values
within a candidate reference section, wherein substantially
one-third of the pixels of said candidate reference sections are
darker than said black-grey threshold and substantially one-third
of the pixels are lighter than said grey-white threshold, and then
rejecting said candidate reference section based on the difference
in values of said black-grey and grey-white thresholds.
16. A method as recited in claim 1 wherein said step of selecting
said reference sections from the most distinctive of said candidate
reference sections includes the steps of: selecting a number of the
most distinctive of said candidate reference sections based on the
results of the autocorrelation calculations, measuring the
redetectability of each of the remaining candidate reference
sections with respect to one or more subsequent two-dimensional
images of said reference object, and selecting said reference
sections according to said degree of similarity between the image
data of said remaining candidate reference sections and the image
data of said one or more subsequent images.
17. A method as recited in claim 16 wherein said step of measuring
the redetectability of each of the remaining candidate reference
sections includes the steps of forming one or more subsequent
two-dimensional images of said reference object and determining the
degree of similarity between the two-dimensional image data of each
of the remaining candidate reference sections and the
two-dimensional image data of said one or more subsequent
images.
18. A method as recited in claim 17 wherein said step of
determining the degree of similarity between the image data of each
of the remaining candidate reference sections and the image data of
said one or more subsequent images includes for each of said
subsequent images the step of determining best-match locations for
said remaining candidate reference sections within corresponding
search areas of said subsequent image, wherein each search area is
a two-dimensional region of said subsequent image that surrounds
the expected location of a corresponding candidate reference
section, wherein each best-match location is the location within a
search area at which the correlation is highest between the image
data of a corresponding candidate reference section and the image
data of the subsequent image, and wherein the correlation value
computed at each best-match location is a measure of said degree of
similarity.
19. A method as recited in claim 18 wherein said step of
determining best-match locations for said remaining candidate
reference sections within corresponding search areas of said
subsequent image includes the steps of first determining the
best-match locations of one of said remaining candidate reference
sections within a first search region of said subsequent image,
wherein the deviation of the best-match location of said one
candidate reference section from the expected location thereof
defines a positional error of said subsequent image with respect to
said reference image, and then determining the best-match locations
of the rest of said remaining candidate reference sections with
respect to said subsequent image by searching within second search
regions for best-match locations, wherein said second search
regions are located with reference to said positional error, and
wherein said second search regions are smaller in size than said
first search region.
20. A method as recited in claim 19 wherein said step of selecting
said reference sections from the remaining candidate reference
sections according to said degree of similarity between the image
data of said reference sections and the image data of said one or
more subsequent images includes the steps of sorting said remaining
candidate reference sections according to the average of the
correlation values computed at the best-match locations for each of
said subsequent images, and then selecting a number of the highest
ranked candidate reference sections as said reference sections.
21. A method as recited in claim 1 further comprising the step of
storing the two-dimensional image and relative positioning data of
each of said reference sections after said step of identifying a
plurality of reference sections for later use in verifying the
identity of objects.
22. A method as recited in claim 1 wherein said step of forming a
two-dimensional verify image includes the steps of first forming an
image consisting of grey pixels having a plurality of brightness
values on a grey scale ranging between black and white, and then
converting said grey pixels into binary pixels having brightness
values of black or white based on the median grey value of
neighboring pixels.
23. A method as recited in claim 1 wherein said step of determining
a best-match location within each verify region includes the steps
of first determining the best-match location of one of said
reference sections within a first verify region of said verify
image, wherein the deviation of the best-match location of said one
reference section from the expected location thereof defines a
positional error of said verify image with respect to said
reference image, and then determining the best-match locations of
the rest of said reference sections with respect to said verify
image by searching within second verify regions for best-match
locations, wherein said second verify regions are located with
reference to said positional error.
24. A method as recited in claim 23 wherein said second verify
regions are smaller in size than said first verify region.
25. A method as recited in claim 23 wherein said step of first
determining the best-match location of said one of said reference
sections within said first verify region includes the steps of
finding the highest correlation value within a subfield of said
first verify region and specifying the best-match location as the
location within said subfield at which said correlation value is
highest if said highest correlation value exceeds a predetermined
value, and if not, finding the highest correlation value with the
entire first verify region and specifying the best-match location
as that location within the entire first verify region at which the
correlation value is highest, wherein each correlation value is a
measure of the similarity between the image data of said one
reference section and the image data of an equal-sized area of said
first verify region.
26. A method as recited in claim 25 wherein said subfield is
located within said first verify region at a location surrounding
the expected location of said one reference section.
27. A method as recited in claim 1 wherein said step of determining
a best-match location includes the steps of computing correlation
values throughout said verify region, wherein each correlation
value is a measure of the similarity between the image data of a
reference section and the image data of an equal-sized area of the
corresponding verify region, and specifying the best-match location
as that location within said verify region at which the correlation
value is highest.
28. A method as recited in claim 27 wherein said step of computing
a correlation value includes the step of comparing pixel-by-pixel
the image data of said reference section and the image data of an
equal-sized area of said verify region.
29. A method as recited in claim 27 wherein said step of
determining a best-match location includes the steps of first
computing correlation values at selected locations distributed
throughout said verify region and then computing correlation values
at all possible locations near those locations where high
correlation values were found.
30. A method as recited in claim 1 wherein said step of forming a
two-dimensional reference image of a reference object includes the
step of forming an image consisting of grey pixels having a
plurality of brightness values on a grey scale ranging between
black and white, wherein said step of identifying a plurality of
reference sections includes the step of converting the grey pixels
of each reference section into binary pixels having brightness
values of black or white based on the median grey value of pixels
within said reference section, wherein said step of forming a
two-dimensional verify image includes the steps of first forming an
image consisting of grey pixels having a plurality of brightness
values on a grey scale ranging between black and white, and then
converting said grey pixels into binary pixels having brightness
values of black or white based on the median grey value of pixels
within said verify image, and wherein said step of determining a
best-match location within each verify region includes the steps of
computing correlation values throughout said verify region, wherein
each correlation value is computed by a pixel-by-pixel comparison
between the binary image data of said reference section and the
binary image data of an equal-sized area of said verify region, and
specifying the best-match location as that location within said
verify region where the correlation value is highest.
31. A method as recited in claim 30 wherein said step of converting
the grey pixels of each reference section into binary pixels
includes the step of trinarizing the image data of a reference
section by first determining a black-grey threshold value and a
grey-white threshold value based on the distribution of grey scale
brightness values within said reference section, and then
converting the grey scale pixels of said reference section into
black, grey, and white pixels according to their grey scale
brightness values relative to said black-grey and grey-white
threshold values.
32. A method as recited in claim 31 wherein said step of computing
a correlation value includes the step of comparing the black pixels
and white pixels of said reference section to the black pixels and
white pixels of said verify region, adding the number of matches
between black or white pixels of said reference section and black
or white pixels of said verify region, and dividing by the number
of black and white pixels of said reference section, wherein the
resulting value equals said correlation value.
33. A method as recited in claim 1 wherein said step of determining
a best-match location includes the steps of computing correlation
values throughout said verify region, wherein each correlation
value is a measure of the similarity between the image data of a
reference section and the image data of an equal-sized area of the
corresponding verify region, and specifying the best-match location
as that location within said verify region where the correlation
value is highest, and wherein said step of verifying the identity
of said object includes the steps of first determining a
displacement value for each of said reference sections, wherein
each displacement value relates to the distance between the
location of a reference segment within said reference image and the
location of the corresponding best-match location within said
verify image, and verifying that said object has the same identity
as said reference object based on high correlation values and low
displacement values.
34. A method as recited in claim 33 wherein said step of verifying
the identity of said object includes the step of adjusting said
displacement values by rotating the best-match locations of the
reference sections around an origin at the best-match location of
one of said reference sections to correct for translational and
rotational misalignment of said verify image with respect to said
reference image.
35. A method as recited in claim 33 wherein said step of verifying
the identity of said object includes the step of establishing a
functional expression utilizing correlation values and displacement
values so that each reference section can be classified as tending
to verify or not depending on its corresponding correlation and
displacement values relative to said functional expression, and
wherein the identity of said object will be verified only when a
predetermined number of said reference sections tend to verify
object identity.
36. A method as recited in claim 35 wherein the identity of said
object will be verified if a majority of said reference sections
tend to verify.
37. A fingerprint verification method for verifying the identity of
a person as recited in claim 1 wherein said step of forming a
reference image of a reference object includes the step of forming
an image of a fingerprint of a person seeking enrollment, and
wherein said step of forming a verify image of an object whose
identity is to be verified includes the step of forming an image of
a fingerprint of a person seeking verification.
38. A fingerprint verification method for verifying the identity of
a person by comparing presently obtained fingerprint data with
previously obtained fingerprint data, said method comprising the
steps of:
first enrolling one or more persons by the steps of forming a
two-dimensional reference image of a fingerprint of each enrolling
person, identifying a plurality of reference sections from each
reference image by partitioning said reference image into a
plurality of two-dimensional candidate reference sections from
which said reference sections will be selected, performing an
autocorrelation calculation for each candidate reference section to
determine the degree of distinctiveness of said candidate reference
section relative to a two-dimensional local region of reference
section by comparing pixels of the two-dimensional image data of
said candidate reference section to corresponding pixels of the
two-dimensional image data of equal-sized areas throughout said
local region, wherein said local region is larger than said
candidate reference section and smaller than said reference image,
and selecting said reference sections from the most distinctive of
said candidate reference sections based on the results f the
autocorrelation calculations, and saving the two-dimensional image
and relative positioning data of each of said reference sections as
fingerprint data for the enrolling person; and then
verifying the identify of someone claiming to be an enrolled person
by the steps of retrieving the two-dimensional image and relative
positioning data of the reference sections of the enrolled person,
forming a two-dimensional verify image of the fingerprint of the
person claiming to be enrolled, wherein said verify image includes
a plurality of two-dimensional verify regions each corresponding in
position to one of said reference sections and each larger in
extent than its corresponding reference section, determining a
best-match location within each verify region at which the image
data is most similar to the image data of its corresponding
reference section, and verifying the identity of the person
claiming to be enrolled according to the degree of similarity
between the two-dimensional image data of said best-match locations
and said corresponding reference sections and according to the
degree of similarity between the relative positioning of said
best-match locations and said corresponding reference sections.
39. A fingerprint verification method for verifying the identity of
a person by comparing presently obtained fingerprint data with
previously obtained fingerprint data, said method comprising the
steps of:
first enrolling one or more persons according to the steps of:
forming a two-dimensional reference image of a fingerprint of a
person seeking enrollment,
partitioning said reference image into a plurality of
two-dimensional candidate reference sections,
determining the degree of distinctiveness of each candidate
reference section within a two-dimensional local region surrounding
said candidate reference section by comparing pixels of the
two-dimensional image data of said candidate reference section to
corresponding pixels of the two-dimensional image data of
equal-sized areas throughout said local region, wherein said local
region is larger than said candidate reference section and smaller
than said reference image,
determining the redetectability of each candidate reference section
with respect to one or more subsequent images of the fingerprint of
the person seeking enrollment by determining the degree of
similarity between the image data of said candidate reference
section and the image data of said one or more subsequent
images,
selecting a group of reference sections from said candidate
reference sections according to the degree of distinctiveness and
the degree of redetectability of said candidate reference sections,
and
saving the two-dimensional image and relative positioning data of
each of the selected reference sections as fingerprint data for
later use in verifying the identity of a person seeking
verification;
and later verifying the identity of a person seeking verification
according to the steps of:
obtaining an indication of which previously enrolled person that
the person seeking verification claims to be,
retrieving the two-dimensional image and relative positioning data
of the reference sections of the previously enrolled person,
forming a two-dimensional verify image of the fingerprint of the
person seeking verification, wherein said verify image includes a
plurality of two-dimensional verify regions each corresponding in
relative position to and each larger in extend that a corresponding
one of said reference sections,
determining the best-match locations of said reference sections
within their corresponding verify regions, wherein each best-match
location is that location within a verify region at which the image
data is most similar to the image data of the corresponding
reference section,
determining a displacement value for each of said reference
sections, wherein each displacement value is a measure of the
distance between the position of the best-match location and the
expected position thereof,
adjusting one or more of said displacement values to correct for
misalignment of said verify image with respect to said reference
image,
classifying each reference section as tending to verify or not
based on its corresponding correlation and displacement values,
and
verifying the identity of the person seeking verification if at
least a predetermined number of said reference sections are
classified as tending to verify.
40. A fingerprint verification method for verifying the identity of
a person by comparing presently obtained fingerprint data with
previously obtained fingerprint data, said method comprising the
steps of:
first enrolling one or more persons according to the steps of:
forming a two-dimensional reference image of a fingerprint of a
person seeking enrollment, wherein said reference image consists of
grey pixels having a plurality of brightness values on a grey scale
ranging between black and white,
partitioning said reference image into a plurality of
two-dimensional candidate reference sections from which a set of
reference sections will be selected,
rejecting certain candidate reference sections based on the grey
scale values of the pixels within said candidate reference
sections,
performing an autocorrelation calculation for each remaining
candidate reference section to determine the degree of
distinctiveness of said candidate reference section within a local
region surrounding said candidate reference section, wherein said
autocorrelation calculation includes the steps of computing a
plurality of correlation values throughout said local region,
wherein each correlation values throughout said local region,
wherein each correlation value is a measure of the similarity
between the image data of said candidate reference section with the
image data of an equal-sized area of said local region, and
computing an autocorrelation score based on the highest correlation
value computed within said local region but outside of a region
near the location of said candidate reference section, wherein a
high autocorrelation score indicates a locally non-distinct
candidate reference section,
selecting a number of the most distinctive of said candidate
reference sections based on the results of the autocorrelation
calculations,
measuring the redetectability of each of the remaining candidate
reference sections with respect to one or more subsequent
two-dimensional images of the fingerprint of the person seeking
enrollment by determining the degree of similarity between the
image data of said remaining candidate reference sections and the
image data of said one or more subsequent images,
selecting a template of reference sections from the set of
remaining candidate reference sections according to said
redetectability of said remaining candidate reference sections with
respect to said one or more subsequent images, and
saving the two-dimensional image and relative positioning data of
each of the selected reference sections as fingerprint data for
later use in verifying the identity of a person seeking
verification;
and later verifying the identity of a person seeking verification
according to the steps of:
obtaining from the person seeking verification an indication of
which previously enrolled person that the person seeking
verification claims to be,
retrieving the two-dimensional image and relative positioning data
of the reference sections of the previously enrolled person,
forming a two-dimensional verify image of the fingerprint of the
person seeking verification, wherein said verify image includes a
plurality of two-dimensional verify regions each corresponding in
relative position to and each larger in extent than a corresponding
one of said reference sections,
determining the best-match locations of said reference sections
within corresponding verify regions of said verify image, wherein
each best-match location is that location within a verify region at
which the image data is most similar to the image data of the
corresponding reference section, wherein said best-match location
is determined by the steps of computing correlation values
throughout said verify region, wherein each correlation value is a
measure of the similarity between the image data of said
corresponding reference section and the image data of an
equal-sized area of said verify region, and specifying the
best-match location as that location within said verify region
where the correlation value is highest,
determining a displacement value for each of said reference
sections, wherein each displacement value is a measure of the
distance between the position of the best-match location and the
expected position thereof,
adjusting one or more of said displacement values to correct for
misalignment of said verify image with respect to said reference
image,
classifying each reference section as tending to verify or not
based on its corresponding correlation and displacement values,
and
verifying the identity of the person seeking verification if at
least a predetermined number of said reference sections are
classified as tending to verify.
41. A method as recited in claim 40 wherein said step of a
performing an autocorrelation calculation includes the steps of
converting the grey pixels of said local region into binary pixels
having brightness values of black or white and converting the grey
scale pixels of said candidate reference section into trinary
pixels having brightness values of black, grey, or white, and
wherein said step of computing correlation values throughout said
local region disregards the grey trinary pixels of said candidate
reference sections.
42. A method as recited in claim 41 wherein each computation of a
correlation value within said local region includes the step of
comparing the black pixels and white pixels of said candidate
reference section to the black pixels and white pixels of said
local region, adding the number of matches between black or white
pixels of said candidate reference section and respective black or
white pixels of said local region, and dividing by the number of
black pixels and white pixels of said candidate reference section,
wherein the resulting value equals said correlation value.
43. A method as recited in claim 40 wherein said method further
includes the step of converting the grey scale pixels of said
reference sections into trinary pixels having brightness values of
black, grey, or white, wherein said step of forming a verify image
includes the step of forming an image with binary pixels having
brightness values of black or white, and wherein said step of
computing correlation values throughout each verify region
disregards the grey trinary pixels of said reference sections.
44. A method as recited in claim 43 wherein each computation of a
correlation value within said verify region includes the step of
comparing the black pixels and white pixels of said reference
section to the black pixels and white pixels of said verify region,
adding the number of matches between black or white pixels of said
reference section and respective black or white pixels of said
verify region, and dividing by the number of black pixels and white
pixels of said reference section, wherein the resulting value
equals said correlation value.
45. A method as recited in claim 40 wherein said step of
determining the best-match locations of said reference sections
within respective verify regions of said verify image includes the
steps of determining a first best-match location of one of said
reference sections within a corresponding first verify region of
said verify image, and then locating the rest of said verify
regions relative to said first best-match location, and then
determining the best-match locations of the rest of said reference
sections.
46. A method as recited in claim 45 wherein said step of adjusting
said displacement values to correct for misalignment of said verify
image includes the step of adjusting for rotational misalignment of
said verify image by finding a rotated position of said template of
reference sections with respect to said verify image at which sum
of the squares of the displacement values is minimized.
47. A fingerprint verification method for verifying the identity of
a person by comparing presently obtained fingerprint data with
previously obtained fingerprint data, said method comprising the
steps of:
first enrolling one or more persons according to the steps of:
forming a reference image of the ridges and valleys of a
fingerprint of a person seeking enrollment, wherein said reference
image consists of grey pixels having a plurality of brightness
values on a grey scale ranging between black and white, and wherein
said fingerprint ridges and valleys are represented by brightness
values tending toward black or white,
partitioning said reference image into a plurality of candidate
reference sections from which a set of reference sections will be
selected,
rejecting certain candidate reference sections based on the grey
scale values of the pixels within said candidate reference
sections,
performing an autocorrelation calculation for each remaining
candidate reference section to determine the degree of
distinctiveness of said candidate reference section within a local
region surrounding said candidate reference section, wherein said
autocorrelation calculation includes the steps of converting the
grey pixels of said local region into binary pixels having
brightness values of black or white, converting the grey scale
pixels of said candidate reference section into trinary pixels
having brightness values of black, grey, or white, computing a
plurality of correlation values throughout said local region,
wherein each correlation value is a measure of the similarity
between the black and white trinary pixels of said candidate
reference section with the black and white binary pixels of an
equal-sized area of said local region, and computing an
autocorrelation score based on the highest correlation value
computed within said local region but outside of a region near the
location of said candidate reference section, wherein a high
autocorrelation score indicates a locally non-distinct candidate
reference section;
selecting a number of the most distinctive of said candidate
reference sections based on the results of the autocorrelation
calculations,
measuring the redetectability of each of the remaining candidate
reference sections with respect to one or more subsequent images of
the finger print of the person seeking enrollment by determining
the degree of similarity between the image data of said remaining
candidate reference sections and the image data of said one or more
subsequent images,
selecting a template of reference sections from the set of
remaining candidate reference sections according to said
redetectability of said remaining candidate reference sections with
respect to said one or more subsequent images, wherein the image
data contained in each of said reference sections is locally
unique, and
saving the image and relative positioning data of each of the
selected reference sections as fingerprint data for later use in
verifying the identity of a person seeking verification;
and later verifying the identity of a person seeking verification
according to the steps of:
obtaining from the person seeking verification an indication of
which previously enrolled person that the person seeking
verification claims to be,
retrieving the image and relative positioning data of the reference
sections of the previously enrolled person,
forming a verify image of the ridges and valleys of the fingerprint
of the person seeking verification, wherein said verify image
includes binary pixels having brightness values of black or white
to indicate said ridges and valleys, and wherein said verify image
includes a plurality of verify regions each corresponding in
relative position to and each larger in extent than a corresponding
one of said reference sections,
determining a first best-match location of one of said reference
sections within a first verify region of said verify image, wherein
each best-match location is that location within a verify region at
which the image data is most similar to the image data of the
corresponding reference section, wherein each best-match location
is determined by the steps of computing correlation values
throughout the verify region and specifying the first best-match
location as that location within the verify region where the
correlation value is highest, wherein each correlation value is a
measure of the similarity between the black and white pixels of the
reference section and the black and white pixels of an equal-sized
area of the corresponding verify region, and then
locating the rest of said verify regions relative to said first
best-match location, and then
determining the best-match locations of the rest of said reference
sections with respect to said verify image by searching within the
rest of said verify regions for best-match locations,
determining a displacement value for each of said reference
sections, wherein each displacement value is a measure of the
distance between the position of the best-match location and the
expected position thereof,
adjusting one or more of said displacement values to correct for
rotational misalignment of said verify image with respect to said
reference image,
classifying each reference section as tending to verify or not
depending on its corresponding correlation and displacement values,
and
verifying the identity of the person seeking verification if at
least a predetermined number of said reference sections are
classified as tending to verify.
48. An apparatus for verifying the identify of a person by
comparing a two-dimensional image of that person's fingerprint to
reference data derived from a two-dimensional fingerprint image
obtained during a prior enrollment procedure, said apparatus
comprising:
means for forming a two-dimensional image of a fingerprint during
both enrollment and verification procedures,
means for generating reference data from a fingerprint image
obtained from a fingerprint during an enrollment procedure, wherein
said reference data includes the two-dimensional image data and
relative positioning of a plurality of reference sections of the
fingerprint image, and wherein the image data contained in each of
said reference sections is distinct relative to the image data
adjacent to and surrounding said reference section and is
redetectable by virtue of its similarity with the image data of one
or more subsequent images of the fingerprint;
means for saving said reference data for later use during the
verification procedure;
means for retrieving the reference data associated with an enrolled
person when a person claiming to be that enrolled person seeks to
be verified;
means for defining a plurality of two-dimensional verify regions in
the fingerprint image of the person seeking verification, wherein
each verify region corresponds in position to one of said reference
sections, and wherein each verify region is larger in extent than
its corresponding reference section;
means for determining a best-match location within each verify
region at which the image data is most similar to the image data of
its corresponding reference section; and
means for verifying the identity of the person claiming to be
enrolled according to the degree of similarity between the image
data of said best-match locations and said corresponding reference
sections, and according to the degree of similarity between the
relative positioning of said best-match locations and said
corresponding reference sections.
Description
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates generally to the field of verification of
object identity by image correlation methods, and relates more
particularly to a method and apparatus for verification of
personnel identity by correlation of fingerprint images.
2. Description of the Relevant Art
Fingerprint matching is a commonly used and well accepted biometric
method of personnel identification. Each fingerprint has a
distinctive pattern of ridges and valleys that makes the
fingerprint unique. The overall ridge patterns of fingerprints can
be classified according to their distinctive shapes into several
classes of morphology, including loops, arches, and whorls. The
individual ridges of fingerprints have distinctive orientations,
spacings, terminations, and bifurcations. Fingerprint matching
methods are based on the premise that the combination of these
features in any one fingerprint is unique.
One use of the fingerprint matching technique is in access control,
wherein personnel are permitted or denied access to a controlled
area based on comparisons with a data base of fingerprints. The
controlled area may be a physical area, in which case access is
controlled by a physical barrier, or a virtual area such as a
computer program or data base, in which case access is controlled
by an electric barrier. The data base of fingerprints is
constructed during an enrollment procedure that consists of
recording in some form the fingerprints of those individuals who
are to be permitted access. Once the data base has been
constructed, an individual will be granted access by way of a
verification procedure only if the fingerprint presented for
verification matches the stored fingerprint data of a particular
enrolled individual.
Since manual methods of fingerprint matching are cumbersome, an
automated method of personnel verification for access control is
desirable. In order to be useful, such an automated method must
accurately verify enrolled personnel, and must also accurately
reject non-enrolled personnel. Inaccuracies in the verification
process have been broken down into two types; a type one error is a
false rejection of an enrolled individual, while a type two error
is a false verification of a non-enrolled individual. Ideally, both
type one and type two errors should be minimized, however,
depending upon the application, an increased rate of one type of
error may be tolerated in order to minimize the rate of the other
type of error. For example, if the automated method is used to
control access to a vault containing highly sensitive documents,
the false verification rate should be very close, if not equal, to
zero in order to protect against unauthorized access, while the
inconveniences associated with a relatively large false rejection
rate can be tolerated. On the other hand, if the cost of a false
rejection is high and the penalty of a false verification is low,
then a relatively high false verification rate can be tolerated in
order to minimize the false rejection rate.
One factor that influences the accuracy of automated methods of
access control is the repeatability of the process of imaging the
fingerprint to be enrolled or verified. As indicated above, the use
of fingerprint matching in access control utilizes two distinct
procedures, enrollment and verification. In a typical automated
method of access control, both the enrollment procedure and the
verification procedure involve forming an optical image of the
fingerprint of the individual to be enrolled or verified. The
process of imaging a fingerprint typically involves sensing light
reflected from the fingerprint, wherein the ridges and valleys of
the fingerprint reflect the light differently. Inaccuracies may
result from the imaging process itself by distortions of the
fingerprint image caused by the imaging apparatus, and may also
result from inconsistencies in alignment of the finger with the
imaging apparatus or in variations of the moisture level of the
finger surface. Another factor that influences the accuracy of
automated methods of fingerprint matching is that the finger itself
may change in size due to physiological or temperature related
causes.
In addition to accuracy, other factors that effect the usefulness
of automated methods of fingerprint matching include the cost of
the automated apparatus, the speed of the enrollment and
verification procedures, and the resistance of the method to
tampering and misuse. Cost and speed are directly influenced by the
efficiency of the enrollment procedure in accurately characterizing
fingerprints by manipulating and storing a minimal amount of
data.
SUMMARY OF THE INVENTION
Broadly stated, the present invention involves an image correlation
method for use in verifying the identity of an object. One aspect
of the method involves an enrollment procedure, and another aspect
of the method involves a verification procedure. More specifically,
the enrollment procedure of the method includes the steps of:
forming a reference image of a reference object; identifying a
plurality of reference sections of the reference image, where the
image data contained in each of the reference sections is distinct
relative to the image data adjacent to and surrounding the
reference section; and saving the image data of each of the
reference sections for later use in the verification procedure.
After performing the enrollment procedure, the verification
procedure can be performed. The verification procedure includes the
steps of: forming a verify image of an object whose identity is to
be verified, where the verify image includes a plurality of verify
regions each corresponding in position to one of the reference
sections and each larger in extent than its corresponding reference
section; determining a best-match location within each verify
region at which the image data is most similar to the image data of
its corresponding reference section; and verifying the identity of
the object according to the degree of similarity between the image
data of the best-match locations and the corresponding reference
sections and according to the degree of similarity between the
relative positioning of the best-match locations and the
corresponding reference sections.
More narrowly stated and in accordance with the illustrated
preferred embodiment, the present invention provides a method and
apparatus for verification of personnel identity by correlation of
fingerprint images. The method includes the steps of: first
enrolling a person by the steps of forming a reference image of a
fingerprint of the person, identifying a plurality of reference
sections within the reference image, where the image data contained
in each of the reference sections is locally unique, and saving the
image data of each of the reference sections; and then verifying
the identity of someone claiming to be an enrolled person by the
steps of retrieving the image data of the reference sections of the
enrolled person, forming a verify image of the fingerprint of the
person claiming to be enrolled, where the verify image includes a
plurality of verify regions each corresponding in relative position
to one of the reference sections and each larger in extent than its
corresponding reference section, determining a best-match location
within each verify region at which the image data is most similar
to the image data of its corresponding reference section, and
verifying the identity of the person claiming to be enrolled
according to the degree of similarity between the image data of the
best-match locations and the corresponding reference sections and
according to the degree of similarity between the relative
positioning of the best-match locations and the corresponding
reference sections.
The apparatus of the present invention consists primarily of an
imaging device for forming the images of the fingerprints, and a
programmed computer and appropriate interface circuits for
performing the tasks of selecting the reference sections, saving
and retrieving the image data contained in the reference sections,
aligning the reference sections on a fingerprint image to be
verified by calculating the best-match locations, and verifying
identity based on the results of the alignment task. In the
preferred embodiment, the imaging device forms a grey scale image
of the fingerprint consisting of a rectangular array of pixels
(picture elements), where each pixel has a digital value ranging
from one value that represents white to another value that
represents black, with values representing shades of grey in
between. Within the computer, for ease of computation, the
fingerprint images are preferably represented in binary form,
namely black and white pixels, with the conversion from grey scale
to binary form being performed either by hardware external to the
computer or by the computer itself.
One aspect of the method for verification of identity by
correlation of fingerprint images includes the definition of the
reference sections that characterize the fingerprint of an enrolled
individual. During the enrollment procedure, the reference
fingerprint image of the person enrolling is analyzed for purposes
of identifying distinctive areas of the fingerprint image. Most
often these distinctive areas correspond to fingerprint features
such as ridge terminations and bifurcations, ridge islands, cores,
deltas, and other differentiating features commonly found in
fingerprints.
The reference fingerprint image is effectively partitioned into a
series of relatively small candidate reference sections, each of
which is analyzed to determine whether it is distinctive enough to
be selected as one of the reference sections. In the preferred
embodiment, for quality control purposes, an initial evaluation is
performed on the grey scale image data of each candidate reference
section. If the candidate reference section is too light, the
candidate reference section is rejected because it is outside the
boundaries of the useful fingerprint image. If the contrast among
the pixels of the candidate reference section is too small, it is
also rejected. Also, the candidate reference section is rejected if
it is too close to the physical edge of the fingerprint image, or
if it is too dark.
After the initial evaluation of candidate reference sections, a
series of calculations are performed in the preferred embodiment to
determine the uniqueness or distinctiveness of each remaining
candidate reference section as compared to an area of the reference
fingerprint image immediately surrounding the candidate reference
section. Preferably, for reasons of computational speed, these
calculations are performed on binary (black and white) images
rather than the grey scale images. The area surrounding the
candidate reference section is converted into a binary image
according to the median grey level within that area. In other
words, all pixels having values darker than the median grey level
are converted into black pixels and all pixels having values
lighter than the median grey level are converted into white pixels.
At the same time, a filtering or smoothing process is performed to
remove noise and smooth edges.
Each candidate reference section is also converted into a binary
equivalent of the grey image, but by another process, identified
herein as "trinarization." In order to eliminate the uncertainty
and variability of edge determinations in a process that converts
grey images into binary images according to the median grey level,
the trinarization technique divides all pixels into one of three
levels, black, grey, and white. A histogram of grey values of the
grey scale image is determined and black-grey and grey-white
threshold values are established according to equal one-third
distributions. All pixels having grey values darker than the
black-grey threshold value are converted into black pixels; all
pixels having grey values lighter than the grey-white threshold
value are converted into white pixels; all other pixels are ignored
in subsequent correlation calculations. Thus, the black and white
pixels represent with high confidence ridge and valley regions of
the fingerprint image, while the grey pixels represent the
transition regions between the ridges and valleys.
Once a candidate reference section has been trinarized and the
surrounding area has been binarized, an autocorrelation calculation
is performed in the preferred embodiment in which the local
distinctiveness of the candidate reference section with respect to
the underlying reference fingerprint image is determined. The term
"autocorrelation" is used herein as referring to a series of
individual correlation calculations each performed between the
trinarized image data of the candidate reference section and the
binarized image data of a subfield of the surrounding area, equal
in size to the candidate reference section, and offset within the
surrounding area by varying amounts. The method of the present
invention assumes that high correlations between the candidate
reference section and other locations within the surrounding area
means that the candidate reference section is not very distinctive
and is, thus, not suitable for use in characterizing the
fingerprint image. Such a high correlation will occur, for example
in cases where the candidate reference section image includes only
parallel ridges without any distinctive fingerprint features. One
would expect a high correlation between such a candidate reference
section and locations within its surrounding area when the
candidate reference section is displaced in a direction parallel to
the ridges or displaced by multiples of a ridge spacing in a
direction perpendicular to the ridges. On the other hand, a
candidate reference section image that includes a ridge termination
may be unique within the surrounding area, in which case only low
correlations will be found. Since high correlations are obtained at
locations at or near the original location of the candidate
reference section, such locations are excluded in the
autocorrelation calculation. The autocorrelation score of each
candidate reference section is the highest correlation value
occurring within the surrounding area, except for the excluded
region near the center of the surrounding area.
Once the autocorrelation calculation has been performed between
each candidate reference section and its corresponding surrounding
area of the reference fingerprint image, the candidate reference
sections are ranked in the preferred embodiment according to their
autocorrelation scores, with low correlation scores being most
desirable because such scores represent locally unique or
distinctive candidate reference sections. A predetermined number of
the most distinctive candidate reference sections are then further
tested against additional reference fingerprint images of the same
person to determine the repeatability of accurately matching the
candidate reference sections to subsequent fingerprint images. The
most distinctive and repeatable candidate reference section is
identified as the primary reference section, which will be used
during the verification procedure to align the reference sections
of the template with respect to subsequent fingerprint images. At
the conclusion of this reference section identification and
selection process, a set of distinctive reference sections is
selected as a "template" that best represents the reference
fingerprint image. The trinarized image data contained in the
reference sections of the template along with data characterizing
their relative positions is then stored, thereby completing the
enrollment procedure.
Another aspect of the method for verification of identity by
correlation of fingerprint images includes the verification
procedure, in which a person seeking access and claiming to be
enrolled provides a fingerprint for comparison with the template of
the enrolled person. In the preferred embodiment, a verify
fingerprint image is obtained from the person seeking access and is
filtered and binarized. The image data defined by the template of
the enrolled person is retrieved from storage and is compared to
the verify fingerprint image.
As an initial step in the verification procedure, according to the
preferred embodiment, the template is aligned with respect to the
verify fingerprint image to cancel out translational misalignment
of the person's finger in the imaging device. In order to align the
template to the verify fingerprint image, the primary reference
section is first located on the verify fingerprint image by
determining a "best-match" location within a search region at which
the correlation between the primary reference section and the
underlying verify fingerprint image is the greatest.
Once the primary reference section is located on the verify
fingerprint image, the remaining reference sections are then
located relative to the best-match location of the primary
reference section. A two dimensional translation correction is
determined according to the best-match location of the primary
reference section, and is used in subsequently locating the
expected positions of the remaining reference sections. For each of
the remaining reference sections, a verify region is defined
centered at the expected position of the reference section. The
verify region is larger in extent than the reference section to
allow for rotational misalignment and dimensional changes of the
finger. A best-match location is determined within each verify
region at that location within the verify region at which the
correlation between the trinarized image data of the reference
section and the binarized image data of the verify fingerprint
image is the highest. The best-match location of each reference
section may occur at the center of the verify region, or may occur
at a location displaced therefrom. Next, a rotational correction is
performed to minimize the displacements of the best-match locations
from their respective expected positions in order to cancel out
rotational misalignment of the verify fingerprint image relative to
the reference fingerprint image obtained during the enrollment
procedure.
Once all of the reference sections have been matched to the verify
fingerprint image and their correlations have been computed, an
evaluation of the correlation values and corrected displacements is
then performed to determine whether to verify or reject the person
seeking access as the enrolled person. Each reference section is
classified as a "hit" or a "miss" according to its correlation
value and corrected displacement. A relatively high correlation
value and a relatively low displacement value are required in order
to be classified as a hit. Finally, if the number or percentage of
hits exceeds a predetermined threshold, the person is verified; if
not, the person is rejected.
Several features of the present invention combine to provide a
verification method with significant advantages over other such
methods known in the prior art. One feature is the care in
selecting the reference sections of the template to ensure an
accurate and repeatable characterization of each fingerprint
presented for enrollment. Another feature is the compact size of
the image data of the reference sections, which permits such data
to be easily and efficiently stored. Still another feature is that,
due to the efficiency of the method and the compactness of the
template data, a fingerprint verification apparatus can be
constructed compactly and inexpensively. Still another feature is
the robustness of the verification procedure, which tolerates
imperfect fingerprint image data caused by misalignment, partial
image, and other factors. A further feature is the speed and
accuracy of the verification procedure, which increases the
usefulness of verification apparatus constructed according to the
present invention. A still further feature is that the underlying
method of object identification need not be limited to only
fingerprint image identification; other useful applications for the
method of the present invention will be obvious to those skilled in
the art of object identification.
The features and advantages described in the specification are not
all inclusive, and particularly, many additional features and
advantages will be apparent to one of ordinary skill in the art in
view of the drawings, specification and claims hereof. Moreover, it
should be noted that the language used in the specification has
been principally selected for readability and instructional
purposes, and may not have been selected to delineate or
circumscribe the inventive subject matter, resort to the claims
being necessary to determine such inventive subject matter.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a perspective view of a fingerprint verification terminal
that incorporates the present invention.
FIG. 2 is a perspective schematic view of a fingerprint imaging
device of the fingerprint verification terminal.
FIG. 3 is a schematic illustration of an exemplary light path
through an optical element of the fingerprint imaging device.
FIG. 4 is an enlarged view of a portion of FIG. 3.
FIG. 5 is a block diagram of the fingerprint verification terminal,
including the fingerprint imaging device, a computer, and
associated interface circuitry.
FIG. 6 is a simplified flow chart of the fingerprint verification
method of the present invention.
FIGS. 7A and 7B are detailed flow charts of an enrollment procedure
of the fingerprint verification method.
FIG. 8 is a schematic illustration of a process of selecting
reference sections during the enrollment procedure.
FIG. 9 is a schematic illustration of a process of trinarizing the
image data of a reference section.
FIG. 10 is a schematic illustration of the process of binarizing
fingerprint image data.
FIG. 11 is a schematic illustration of a reference section and an
area surrounding the reference section within which an
autocorrelation calculation is performed.
FIGS. 12A through 12C are schematic illustrations of the results of
the autocorrelation calculation.
FIG. 13 is a schematic illustration of an exemplary reference
fingerprint image, including selected reference sections.
FIG. 14 is a detailed flow chart of a verification procedure of the
fingerprint verification method.
FIG. 15 is a schematic illustration of a fingerprint image to which
a primary reference section and two secondary reference sections
are aligned.
FIGS. 16A through 16C are schematic illustrations of a rotational
correction accomplished during the verification procedure.
FIG. 17 is a schematic plot of correlation values versus corrected
displacement values determined during the verification procedure
and used as criteria for verification and rejection.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
FIGS. 1 through 17 of the drawings depict various preferred
embodiments of the present invention for purposes of illustration
only. One skilled in the art will readily recognize from the
following discussion that alternative embodiments of the structures
and methods illustrated herein may be employed without departing
from the principles of the invention described herein.
The preferred embodiments of the present invention are a method and
an apparatus for verification of personnel identity by correlation
of fingerprint images. As shown in FIG. 1, a fingerprint
verification terminal 10 incorporates the method and apparatus of
the present invention. The verification terminal 10 includes a
housing 12 that encloses the fingerprint verification apparatus of
the present invention. The housing 12 includes an inclined front
panel 14 on which a keyboard 16 and a display 18 are mounted. The
front panel 14 further includes a recess in which an optical
element 20 of an imaging device is mounted. The preferred
embodiment of the apparatus of the present invention incorporates
certain aspects of an imaging device as disclosed in U.S. Pat. No.
4,537,484, issued Aug. 27, 1985 to Fowler, et al., which is herein
incorporated by reference.
The verification terminal 10 is used for both enrollment and
verification. If the individual has been previously enrolled, the
individual is requested to enter his personal identification
number, name, or other form of identification. Then, the individual
places a finger 22 or thumb in the recess defined by the optical
element 20. Sensing circuitry of the verification terminal 10
detects the presence of the digit on the element, which causes a
series of prompting messages to be shown on the display 18. The
finger 22 is then scanned utilizing the imaging device to provide
an image that corresponds to the individual's fingerprint. Next, a
comparison is performed between the present fingerprint image and
fingerprint image data previously defined and stored during
enrollment. If the comparison is positive, the individual's
identity is confirmed and he is granted access. If the comparison
is negative, the individual is not granted access.
The enrollment procedure is similar to the verification procedure
in that the individual's finger is optically scanned and image data
corresponding to the fingerprint image is processed and stored in a
non-volatile memory. Also, identification information, such as the
individual's name or personal identification number, is entered
into the terminal 10 via the keyboard 16. Security personnel are
typically present during the enrollment procedure to verify the
individual's identity. Once the individual has been enrolled,
physical or virtual access may be obtained by the individual
utilizing the previously-described verification procedure, during
which time security personnel need not be present.
Details of the imaging device 52 are seen in FIG. 2 to include the
optical element 20, a light source 24, mirrors 26 and 28, lens 30,
and a light sensor 32. The elements 24 through 32 are mounted on a
carriage 34, which is rotatably positioned about an axis 36 by a
stepper motor 38. The optical element 20 is stationary, and its
curved surfaces are polished so as not to disperse light passing
therethrough. Preferably, the curved surfaces of the optical
element 20 have axes that are coincident with the carriage rotation
axis 36.
As shown in FIGS. 3 and 4, light rays from the light source 24
enter the optical element 20, some of which are reflected from the
upper curved surface 40 thereof back to the sensor 32 by way of the
mirrors 26 and 28 and the lens 30. At locations on surface 40 where
the finger is not in contact due to the presence of a valley 42 in
the fingerprint, the incident light 44 is totally reflected to the
sensor 32. However, at locations on the surface 40 where ridges 46
of the fingerprint are in contact with the surface 40, most of the
incident light 48 is absorbed by the finger 22 instead of being
totally reflected to the sensor 32.
At any one rotational orientation of the carriage, the sensor 32 is
scanned to create a one-dimensional image of a section of the
fingerprint parallel to the axis 36. In order to create a
two-dimensional image of the fingerprint, the carriage 34 is
rotated throughout an arc and the sensor 32 is periodically scanned
until the fingerprint image has been completely formed.
The hardware of the fingerprint verification terminal 10 is seen in
FIG. 5 to include a programmed computer 50 that is utilized as the
control and data processing hub of the terminal. Due to the
simplicity of the calculations required by the verification method,
the computer 50 can be a microprocessor based microcomputer. In
addition to the above-described elements, the imaging device 52
also includes a position detector 54 that is coupled to the stepper
motor 38 for detecting the rotational position of the stepper motor
and the carriage 34. The imaging device 52 also includes a lamp,
motor, and detector control circuit 56, which is coupled to the
computer 50 via a control interface 58. The control circuit 56,
under computer control, turns on the light source 24 and directs
the stepper motor 38 to rotate the carriage when an fingerprint
image is needed. The position detector 54 informs the computer 50
of the rotary position of the motor 38 and carriage 34 so that the
light sensor can be scanned at the appropriate times to create the
two-dimensional video image of the fingerprint. The computer 50
receives fingerprint images from the light sensor 32 through a
video interface circuit 60, which digitizes the analog signals
generated by the light sensor. Alternatively, the video signal can
be routed from the video interface circuit 60 to the computer 50
through a filter circuit 62 and a binarizer circuit 64. The filter
circuit 62 preferably utilizes a LAPLACIAN filtering technique to
smooth the edges and remove noise from the image data. The
binarizer circuit 64 converts the grey-scale image data supplied by
the video interface 60, pixel by pixel, into binary image data,
black and white, preferably according to the local median grey
level of the incoming pixels.
In addition to the imaging device 52, other peripherals are coupled
to the computer 50. A memory 66 is coupled to the computer 50 and
is utilized as a scratch pad memory during the enrollment and
verification procedures. The memory 66 may also include a
non-volatile portion for use in storing the program executed by the
computer 50, and also storing the fingerprint enrollment data,
which is generated during the enrollment procedure and which is
retrieved during the verification procedure. Other storage devices
for storage of the fingerprint enrollment data can be accessed
through an input/output interface circuit 68. Smart cards, which
are credit card sized devices with on-board processing and memory
capability, can be accessed through a smart card interface circuit
70 for use in storing the fingerprint enrollment data of any
particular individual. In such case, the individual can carry with
him at all times his particular enrollment data, which could be
called upon to provide access through verification terminals
without local data storage. Other storage devices such as disk
drives or other computers can be accessed through an RS-232 port 72
or a network driver circuit 74.
The computer also interfaces with other input and output devices,
including the keypad 16 and the display 18, which are located on
the face of the fingerprint verification terminal 10 and are used
to interact with the user of the terminal. A door relay 76 can be
actuated by the computer 50 through the input/output interface 68
to provide means for allowing physical access to an individual upon
the successful conclusion of a verification procedure.
Having thus described the fingerprint verification apparatus of the
present invention, the fingerprint verification method of the
present invention will now be explained in relation to FIGS. 6-17.
As shown in FIG. 6, the fingerprint verification method can be
divided into two procedures, enrollment and verification. The
purpose of the enrollment procedure is to characterize the
fingerprint of a person being enrolled, and to store certain
information about the fingerprint for later use in the verification
procedure. As a first step in the enrollment procedure, the
individual seeking enrollment places his finger 22 on the optical
element 20 and instructs the fingerprint verification terminal 10
to begin the enrollment process. The imaging device 52 of the
fingerprint verification terminal 10 first captures an image of the
individual's fingerprint, labeled in FIG. 6 as a reference image.
As described above, the reference fingerprint image consists of a
two dimensional array of pixels (picture elements) that is
generated by the imaging device 52. The computer, under program
control, subdivides the image into relatively small areas, denoted
herein as candidate reference sections. Each candidate reference
section is analyzed to determine its level of local distinctiveness
and repeatability, and the most distinct and repeatable reference
sections are selected for use in characterizing that fingerprint.
This group of reference sections, denoted herein as a template,
contains small pieces of the image data of the reference
fingerprint image, each containing a locally unique landmark
feature. At the conclusion of the enrollment procedure, the
template is stored in some form of a non-volatile memory, such as a
smart card 78.
The second half of the fingerprint verification method involves the
verification procedure, the purpose of which is to verify or reject
a person seeking access based on a comparison of that person's
fingerprint with the template of whomever that person claims he is.
During the verification procedure, the person seeking access
identifies himself by keying in his name or personal identification
number or some other indication of his identity. The computer,
under program control, retrieves the template of the enrollee
identified by the person seeking access for use in verifying
whether of not that person is the enrollee. The computer also
directs the imaging device 52 to capture a verify fingerprint image
from the person seeking access. Once the verify fingerprint image
has been obtained, the next step of the method is to register the
template with respect to the verify fingerprint image and to
determine how closely the image data contained in the template
matches the corresponding image data of the verify fingerprint.
Based on the results of this determination, a decision is then made
as to whether the person seeking access is or is not the same
person as the enrollee. If the person's identity is verified, then
access is permitted; if not, access is denied.
With the above abbreviated description of the method of the present
invention in mind, the method will now be described in greater
detail. FIGS. 7A and 7B illustrate the flow chart of the program
executed by the computer 50 during the enrollment procedure, while
FIG. 8 schematically illustrates the enrollment procedure. The
first step in enrolling an individual is to capture a fingerprint,
the first reference fingerprint image. To do so, the computer
instructs the imaging device 52 to illuminate the individual's
finger 22, to rotate the carriage 34 and attached imaging
apparatus, and to generate a two-dimensional pixel image, in grey
tones, of the individual's fingerprint. During the initial stages
of the enrollment procedure, the computer 50 uses the actual grey
values instead of the binary equivalents, thus, the video data is
in this case supplied directly to the computer, bypassing the
filter and binarizer circuits 62 and 64.
Once the reference fingerprint image 80 (FIG. 8) has been captured,
the fingerprint image is divided into candidate reference sections
82, each of which is evaluated for possible inclusion in the
template. Each candidate reference section 82 is small in area
compared to the entire reference fingerprint image. In the
preferred embodiment, for example, the size of the image area is
384 by 384 pixels, while the size of each candidate reference
section is 32 by 32 pixels. In order to ensure that distinctive
areas of the reference fingerprint image are not missed, adjacent
candidate reference sections overlap by 16 pixels.
One by one, each of the candidate reference sections is evaluated
according to several criteria. At the beginning of a program loop,
a LAPLACIAN filtering calculation is performed to remove noise and
smooth edges. Then, the grey level values of a candidate reference
section is copied into a working area of the memory, and various
statistical parameters are calculated. A histogram of the grey
value distribution of the pixels of the candidate reference section
is created and the median, one-third, and two-thirds grey threshold
values are determined. If the median grey level is too bright or
too dark, then the candidate reference section is rejected. The
bottom row of candidate reference sections in the reference
fingerprint image 80 would be rejected by this test. Next, if the
range between the one-third and two-thirds threshold values is too
narrow, then the contrast is poor and the candidate reference
section is rejected. The candidate reference section is not
considered further if it is too close to the edge of the physical
fingerprint image.
If the candidate reference section survives these initial tests,
then the grey scale image data is "trinarized" to eliminate from
succeeding calculations those pixels having grey scale values in
the middle third of the distribution. This trinarization process
allows the subsequent correlation calculations to be performed
using binary arithmetic, while eliminating the uncertainty and
variability caused by edge placements in converting the grey scale
image data into binary image data. As shown in FIG. 9, the pixels
of the candidate reference section 84 can be thought of as having
either a black value, a grey value, or a white value, depending
upon their grey scale values relative to the one-third and
two-thirds thresholds. During the trinarization process, only white
and black pixels of the candidate reference section 84 are used in
subsequent calculations; the grey pixels are effectively discarded.
Two new images, a ridge section 86 and a valley section 88, are
generated by the trinarization process. Assuming that black pixels
represent ridges of the fingerprint, then all pixels in the black
third of the grey scale distribution are given black values (1 or
true) in the ridge section 86 and all others are given white values
(0 or false). Also, all pixels in the white third of the grey scale
distribution (valleys) are given black values (1 or true) in the
valley section 88 and all others are given white values (0 or
false). Thus, the black (true) pixels of the ridge section 86
represent ridges and the black (true) pixels of the valley section
88 represent valleys. The resulting binary data contained in the
ridge and valley sections 86 and 88 represent those areas of the
reference fingerprint image in which ridges and valleys will be
consistently found, regardless of edge variability.
With reference now back to FIG. 7A, the next step after the
trinarization is to test whether the now trinarized ridge and
valley section images 86 and 88 contain enough black pixels.
Ideally, each should contain about one third black pixels, but if
the number of pixels is below a threshold of, for example, twenty
percent, the candidate reference section is rejected and the loop
continues with the next adjacent candidate reference section. If
the candidate reference section passes this test, then the location
of the candidate reference section is saved and an area of the
reference fingerprint image surrounding the location of the
candidate reference section is binarized. During this binarization
by the programmed computer, an area of the reference fingerprint
image surrounding the candidate reference section is converted into
a binary image according to the median grey level within the area.
In other words, all pixels having values darker than the local
median grey level are converted into black pixels and all pixels
having values lighter than the median grey level are converted into
white pixels. FIG. 10 schematically illustrates the conversion
through binarization of a grey level image 90 into a binary image
92.
The area binarized is larger than the candidate reference section
because the next step is to determine the local uniqueness of the
candidate reference section within its surrounding area. In the
preferred embodiment, as illustrated in FIG. 11, the candidate
reference section 94 is 32 by 32 pixels in size and the surrounding
area 96 is 64 by 64 pixels in size. Once a candidate reference
section has been trinarized and the surrounding area has been
binarized, an autocorrelation calculation is performed in which the
distinctiveness of the candidate reference section within its
surrounding area is determined. The term "autocorrelation" is used
herein as referring to a series of individual correlation
calculations each performed between the trinarized image data of
the candidate reference section and the binarized image data of a
subfield of the surrounding area equal in size to the candidate
reference section and offset within the surrounding area by varying
amounts. This calculation is labeled "auto" because it involves
correlations between portions of the same image data.
Each correlation calculation results in an autocorrelation score
that indicates the degree of similarity between the candidate
reference section and the subfield of the reference fingerprint
image. Since the image data of both the trinarized candidate
reference section and the binarized subfield are binary values, the
correlation calculation is fairly simple, which results in a rapid
determination of the measure of uniqueness. The formula is:
where CV is the correlation value, Sum is a summation operation
over all of the pixels of the candidate reference section (e.g., 32
by 32), R is the binary value (1 or 0) of the trinarized ridge
section 86 of the candidate reference section, V is the binary
value (1 or 0) of the trinarized valley section 88 of the candidate
reference section, and S is the binary value (1 or 0) of the
subfield of the surrounding area 96. In effect, this calculation
counts the number of times that a ridge (R=1 and S=1) occurs in
both the candidate reference section and the subfield plus the
number of times that a valley occurs in both the candidate
reference section and the subfield (V=1 and S=0), divided by the
total number of ridge and valley pixels in the candidate reference
section. If the correlation value is equal to one, then there is a
perfect match between the image data of the candidate reference
section and the subfield; if the correlation value is equal to
zero, then there is no relationship between the image data of the
candidate reference section and the subfield.
The autocorrelation routine seeks to determine how unique the
candidate reference section is with respect to its surrounding
area. A correlation value is calculated for nearly every possible
location of the candidate reference section within the surrounding
area. The higher the correlation value, the more similar the
candidate reference section is to that location of the surrounding
area. Regardless of the content of the candidate reference section,
the correlation value with the subfield located in the center of
the surrounding area will, by definition, be equal to one. High
correlation values will also be obtained within a few pixels offset
from the center location due to the elimination of the medium grey
pixels from the ridge and valley sections 86 and 88 during the
trinarization process. Accordingly, the locations at which high
correlation values are expected are eliminated from the
autocorrelation calculation by performing the correlation
calculation only at locations outside of a small radius from the
center location. In the preferred embodiment with 32 by 32
reference sections and 64 by 64 surrounding areas, an area having a
6 pixel radius is utilized as an exclusion area.
The uniqueness or distinctiveness of a candidate reference section
is a function of the maximum correlation value outside of the
central exclusion area. A very distinctive or unique candidate
reference section having, for example, a ridge bifurcation, such as
candidate reference section 94 in FIG. 11, will have high
correlation scores only near the center of the surrounding area 96
and within the exclusion area. This result is shown in FIG. 12A,
wherein the only high correlation values are to be found at the
center of the surrounding area 96 and low correlation values are
found elsewhere. The reason that the candidate reference section 94
has a low correlation outside the central exclusion area is that no
other ridge bifurcations exist within the surrounding area.
If, on the other hand, the candidate reference section contains
only parallel ridges and valleys, the result of the autocorrelation
process would be a map of correlation values as shown in FIG. 12C.
Note that fairly high correlation values are obtained at multiples
of the ridge spacing from the central locations because the ridges
of the candidate reference section are not very unique within the
surrounding area. Thus, candidate reference sections having high
correlation values outside the central exclusion area denote poor
choices for the template, while candidate reference sections having
uniformly low correlation values outside the central exclusion area
are good choices for the template.
The result of the autocorrelation step is a number, the
autocorrelation score, which is equal to the highest correlation
value found outside the central exclusion area. A low
autocorrelation score indicates a locally unique candidate
reference section, while a high autocorrelation score indicates a
candidate reference section that is not particularly unique.
In reference now to FIG. 7A, the next step in the process after the
autocorrelation calculation is to loop back and perform the same
series of steps on the next adjacent candidate reference section.
Once all of the candidate reference sections have been evaluated,
the number of candidate reference sections remaining under
consideration is tested to determine whether enough remain. If not,
the enrollment procedure terminates with a bad enrollment indicated
to the individual seeking enrollment. If enough candidate reference
sections remain, then they are sorted according to their
autocorrelation scores, with lower autocorrelation scores denoting
the most unique and desirable candidate reference sections. Next,
the list of candidate reference sections is scanned to determine
whether any two adjacent candidate reference sections remain on the
list, and, if so, the less unique candidate reference section is
discarded. An overlap of fifty percent (adjacent candidate
reference sections) is not permitted, while an overlap of
twenty-five percent (diagonal candidate reference sections) is
permitted. At this point, candidate reference sections may also be
rejected as too close to a too-white (too bright) or too-grey (poor
contrast) region of the reference fingerprint image. If the
candidate reference section does not overlap a more unique adjacent
candidate reference section, then the image data for that candidate
reference section is stored in memory. Once the trinarized image
data and positional coordinates of all of the acceptable candidate
reference sections have been stored, then the number is again
tested. If enough acceptable candidate reference sections remain,
e.g., twenty-five, the enrollment procedure continues, as indicated
on FIG. 7B. At this point, several candidate reference sections 98
remain under consideration, as shown in FIG. 8. The most unique
candidate reference section, namely, the candidate reference
section with the lowest autocorrelation score, will be referred to
herein as the primary candidate reference section, and all other
candidate reference sections as secondary candidate reference
sections.
The enrollment procedure up to this point has been concerned with
selecting unique reference sections of the first reference
fingerprint image. The next step is to investigate the
repeatability of those candidate reference sections selected from
the first reference fingerprint image by looking at additional
reference fingerprint images. Accordingly, the individual seeking
enrollment is instructed to present the same finger so that another
reference fingerprint image can be obtained. Once the second
reference fingerprint image is captured, the image data is filtered
and binarized, this time by the filter and binarizer circuits 62
and 64. Preferably, the filter 62 and binarizer 64 perform their
respective operations on the incoming image data in the same manner
in which the programmed computer 50 filters and binarizes the first
reference fingerprint image internally during the first portion of
the enrollment procedure. Next, the primary candidate reference
section is retrieved from memory and is used to align the secondary
candidate reference sections with the second reference fingerprint
image 100 (FIG. 8). At this point, a search is performed to find
the best-match location for the primary candidate reference section
102 within a search area 104. The search area is centered at the
coordinates of the primary candidate reference section 102 as
determined from the first reference fingerprint image. The size of
the search area 104 is large enough to accommodate a certain amount
of misregistration of the finger 22 with respect to the optical
element 20. Correlation calculations are performed throughout the
search area and the best-match location of the primary candidate
reference section 102 within the search area 104 defines an
alignment correction to be applied to locating the secondary
candidate reference sections.
Preferably, the best-match location is determined by a two-step
process utilizing first a coarse grid, and then utilizing a fine
grid. Initially, the entire search area is covered, computing
correlation values at, for example, every third location. Then, the
correlation calculation is performed for every location surrounding
the few best locations found initially. The time for finding the
best-match location is thus reduced.
As shown in FIG. 8, the second reference fingerprint image 100 is
fairly well aligned with the first reference fingerprint image 80,
and, as a result, the best-match location for the primary candidate
reference section 102 is in the middle of the search area 104. The
third reference fingerprint image 106, however, is shifted slightly
to the right with respect to the first reference fingerprint image
80, and, as a result, the best-match location for the primary
candidate reference section 102 is also shifted slightly toward the
right side of the search area 104. The fourth reference fingerprint
image 108 is shifted slightly downward with respect to the first
reference fingerprint image 80, thus, the best-match location for
the primary candidate reference section 102 is also shifted
downward.
Once the best-match location and its corresponding correlation
value for the primary candidate reference section is determined,
all of the secondary candidate reference sections are matched to
the reference fingerprint image. Each search area 106 for the
secondary candidate reference sections 108 is located relative to
the best-match location of the primary candidate reference section
102 according to the known relative position between the secondary
candidate reference section and the primary candidate reference
section. Thus, the alignment correction determined by the primary
candidate reference section is used to center the search areas for
the secondary candidate reference sections near the expected
best-match locations. The search areas for the secondary candidate
reference sections can be smaller in size than the search area for
the primary candidate reference section as a result of the
alignment correction. In the preferred embodiment, for example, the
search area for the primary candidate reference section is 160 by
192 pixels, while the search areas for the secondary candidate
reference sections are 64 by 64 pixels. For each secondary
candidate reference section, a correlation value is computed at its
best-match location within its corresponding search area, and that
value is stored for later use.
Once the best-match locations and corresponding best correlation
values have been computed for all candidate reference sections,
another reference fingerprint image is obtained from the individual
seeking enrollment, and the correlation loop is repeated. After the
correlation loop has been completed for all of the additional
reference fingerprint images, totalling three in the example shown
in FIG. 8, an evaluation is performed on the candidate reference
sections. Each candidate reference section is ranked according to
the mean of its best correlation values as determined by the above
redetection process. If, for example, three reference fingerprint
images are obtained in addition to the initial reference
fingerprint image, then each candidate reference section would have
three corresponding best correlation values to be averaged. The
highest ranked candidate reference sections at this point are then
selected as the reference sections 109 for inclusion into the
template that represents the culmination of the enrollment
procedure. In the preferred embodiment, the nine highest ranked
candidate reference sections 109 comprise the template. The
reference section with the highest redetection score, referred to
hereinafter as the primary reference section, will be utilized
during the verification procedure in much the same manner as the
primary candidate reference section is utilized in the redetection
process.
One additional test is performed to make sure that the correlation
values are at least equal to a predetermined minimum value. This
test ensures that the enrollee does not switch fingers after the
initial reference fingerprint image is obtained. At this point, the
selected reference sections should be fairly redetectible, so that
a threshold of perhaps 80% average correlation would be reasonable
to use. Once this test is passed, the trinarized image data of the
selected reference sections along with data characterizing their
relative positions is then stored in a non-volatile memory for
later use in the verification procedure. FIG. 13 illustrates an
exemplary fingerprint image with nine reference sections that form
the template for that fingerprint.
The above-described enrollment procedure thus selects based on a
wide coverage and unbiased location, a set of locally unique
reference sections for characterizing the fingerprint of the
enrollee. These reference sections are also repeatable, as assured
by the relocation routine, which simulates the actual verification
procedure described below.
Now that the individual is enrolled, he can at a later time seek
access by participating in the verification procedure of the
present invention. The verification procedure, illustrated in FIG.
14, begins with the individual identifying himself to the terminal
by way of a name or personal identification number, or perhaps by
inserting a smart card containing the stored template data into the
smart card interface device 70. Based on the individual's claimed
identity, the terminal retrieves the template data for use in
verification. The individual then places his finger 22 on the
optical element 20 of the fingerprint verification terminal 10 and
instructs the terminal to verify his identity. The computer 50
commands the imaging device 52 to obtain a verify fingerprint image
of the individual seeking verification. The verify fingerprint
image is filtered and binarized by the filter and binarizer
circuits 62 and 64 prior to arrival at the computer 50.
Once the computer 50 has obtained the verify fingerprint image and
retrieved the template data, the template is aligned with respect
to the verify fingerprint image to cancel out translational
misalignment of the individual's finger in the imaging device. In
order to align the template to the verify fingerprint image, the
primary reference section 110 (FIG. 15) is first located on the
verify fingerprint image by determining the best-match location
within a relatively wide search region 112 at which the correlation
between the primary reference section and the underlying verify
fingerprint image is the greatest. As described above, the
best-match location is determined using the trinarized ridge and
valley section image data of the primary reference section and the
binarized image data of the verify fingerprint image, using the
above disclosed formula for correlation value, and, preferably,
using a coarse/fine grid locate routine, as described above.
Once the best-match location of the primary reference section has
been determined, the remaining reference sections of the template
are then located on the verify fingerprint image. A two-dimensional
alignment correction is determined and is used in locating the
search regions for the remaining reference sections. For each of
the remaining reference sections 114, a verify region 116 is
defined centered at the expected position of the reference section
relative to the position of the primary reference section. The
verify region is larger in extent than its corresponding reference
section to allow for rotational misalignment and dimensional
changes to the finger due to swelling and shrinkage. A best-match
location is determined within the verify region surrounding the
registered position of the reference section. The best-match
location is defined as that location within the verify region at
which the correlation value between the trinarized image data of
the reference section and the binarized image data of the verify
fingerprint image is the highest. The best-match location of each
reference section may occur at the center of the verify region, or
may occur at a location displaced therefrom. For each reference
section, two values are stored: the correlation value calculated at
the best-match location, and the displacement of the best-match
location from the expected position of the reference section.
Once all of the reference sections of the template have been
matched to the verify fingerprint image, a rotational correction is
performed to cancel out rotational misalignment of the verify
fingerprint image relative to the reference fingerprint image
obtained during the enrollment procedure. Preferably, a least
squares fit routine is executed to determine a rotation of the
reference sections that results in a minimized set of displacements
from their expected locations.
The rotational correction process is illustrated in FIG. 16. A
primary reference section 120 has been located within its search
region and the verify regions 124 of the remaining reference
sections 122 have been defined. If the verify fingerprint image
happens to be perfectly aligned with the reference fingerprint
image, then no rotational correction is needed, and the best-match
locations for the reference sections 122 would be located as shown
in FIG. 16A. On the other hand, if the verify fingerprint image is
rotated with respect to the reference fingerprint image, then a
rotational correction is needed, as shown in FIG. 16B. Even though
all of the reference sections 122 of FIG. 16B are shifted from
their expected positions at the centers of their respective verify
regions 124, the rotational correction would cancel out the
apparent misalignment.
While FIGS. 16A and 16B assume that the person seeking verification
is the same as the person who enrolled, FIG. 16C illustrates the
expected result from an impostor. Once the primary reference
section 120 is located at its best-match location, the remaining
reference sections 122 are located at their respective best-match
locations. Since the person verify fingerprint image is from an
impostor, the best-match locations of the reference sections 122
will occur at random displacements from the expected locations in
the centers of the verify regions 124. In this case, the rotational
correction would not significantly reduce the accumulated
displacement errors.
Once all of the reference sections have been located and
rotationally corrected and their correlation values have been
computed, an evaluation of the correlation values and corrected
displacements is then performed to determine whether to verify or
reject the person seeking access as the enrolled person. Each
reference section is classified as a "hit" or a "miss" according to
its correlation value and corrected displacement. A relatively high
correlation value and a relatively low displacement are required in
order for that reference section to be classified as a hit. If the
person seeking verification is the same as the enrollee, then high
correlation values and low displacements would be expected. On the
other hand, if the person seeking verification is an impostor, low
correlation values and relatively high displacement would be
expected. A shown in FIG. 17, a line 126 can be established as a
function of the correlation value and corrected displacement to
define misses above the line and hits below the line. Preferably,
if the number of hits exceeds a predetermined threshold, such as
one half of the number of reference sections in the template, then
the verification is successful, if not, then the person is
rejected.
As shown in FIG. 17, there may be a slight overlap between the
envelope of true cases and the envelope of impostors. If type 2
errors are to be avoided at all costs (false verifications) then
the line 126 can be lowered to a point outside the impostor
envelope by tightening up on the required correlation and
displacement values for a "hit" or by increasing the percentage of
hits required to verify. By proper selection of the hit-miss
criteria, the method of the present invention can reduce the type 2
errors to approximately zero. Type 1 errors (false rejections) can
be minimized by proper placement of fingers in the fingerprint
verification terminal 10 by the users thereof. Note that the
envelopes of FIG. 17 do not indicate the probability distribution
of the true case, which are heavily weighted toward high
correlation and low displacements. It is unlikely that a majority
of the reference sections of a true case would occur in the overlap
area.
One alternative to the above described routine for the
determination of the best-match location of the primary reference
section involves the use of an expandable search area in order to
speed up the verification process for those individuals who
accurately position their fingers in the fingerprint verification
terminal 10. According to this alternative, the best-match search
is first performed within a substantially reduced area surrounding
the expected position of the primary reference section. If within
this reduced area a very high correlation value is calculated, then
that location is assumed to be the best-match location without
searching the entire search region. If the best correlation value
is not high enough then the search area can be expanded in stages
until the best-match location is found. The advantage of this
alternative is that it rewards accurate finger placement by
speeding up the verification process.
Other alternatives exist that depart in various ways from the above
described preferred embodiment. For example, instead of fixing the
reference sections on a fixed grid, locally better reference
sections could be obtaining by examining the regions surrounding
reference sections determined from a fixed grid. In the selection
of the number of reference sections utilized, there are trade-offs
between higher accuracy with an increased number verses increased
computational time. Another alternative is to rotationally correct
based on a subset of the reference sections, so that any one or a
few poorly located reference sections would not pollute the
correction. Also, the image data of the reference sections could be
binarized, while the image data of the reference and verify
fingerprint images could be trinarized. In addition, other
correlation formulas for the measurement of the comparison between
two data sets could be utilized.
From the above description, it will be apparent that the invention
disclosed herein provides a novel and advantageous method and
apparatus for the verification of object identity by image
correlation, and more specifically provides a method and apparatus
for verification of personnel identity by correlation of
fingerprint images. The foregoing discussion discloses and
describes merely exemplary methods and embodiments of the present
invention. As will be understood by those familiar with the art,
the invention may be embodied in other specific forms without
departing from the spirit or essential characteristics thereof. For
example, the object identification method could be used to identify
objects other than fingerprints. Accordingly, the disclosure of the
present invention is intended to be illustrative, but not limiting,
of the scope of the invention, which is set forth in the following
claims.
* * * * *